Predicting the risk for corneal graft rejection by aqueous humor analysis

Abstract

Purpose: Cytokine patterns determined in the aqueous humor before penetrating keratoplasty (PK) may enable us to predict immune reactions (IR). We therefore analyzed 6 cytokines in the aqueous humor of patients before PK. By prospective clinical follow-up, we tested whether patients who developed an IR would present different preoperative cytokine patterns compared to patients without IR.

Methods: We analyzed 18 samples of aqueous humor from 18 patients undergoing PK. The following cytokines were analyzed by cytometric bead array: interleukin 2 (IL-2), interleukin 4 (IL-4), interleukin 5 (IL-5), interleukin 10 (IL-10), tumor-necrosis-factor α (TNF-α), and interferon γ (INF-γ). Seven patients presented with signs of IR during follow up. We performed Cox proportional hazards analysis to determine significant predictors for IR. We iteratively eliminated all co-variates with p values over 0.1 from the survival model (backward selection).

Results: Our final Cox model included the hazardous factors IL-4 (p=0.043) and INF-γ (p=0.059), protective factors IL-2 (p=0.081), IL-5 (p=0.028), and age at time of surgery (p=0.029). We performed a linear discriminant analysis based on these coefficients. The resulting function was: (-9.979*IL5) + (9.262*IL4) + (-3.928*IL2) + (1.709*IFN-γ) + (-0.183*age). A median of -4.97 separated patients with and without IR with no classification error.

Conclusions: We demonstrate that cytokine levels in the aqueous humor can be predictive for IR. Our method allowed an almost 100% separation between patients with and without IR. This finding has the potential to improve the aftercare of PK fundamentally. However, our results need to be confirmed in a larger prospective cohort.

Figure 1

Backward selection. As IL10 showed the least significant result (p=0.78) in the first analysis (A), it was excluded from the second analysis. In the second analysis (B), TNF-α showed the least significant result (p=0.80), and was excluded from the third analysis (C). As the level of significance for backward selection was chosen to be p<0.1, all factors included in analysis three (age, IL2, IL4, IL5, and IFN-γ) were included in the final Kaplan–Meier survival analysis.

Figure 2

Occurence of immune reactions separated by the median of the cytokine score. Kaplan–Meier survival analysis showing an almost 100% separation accuracy between patients with and without immune reactions following PK by the median (−4.97) of the calculated cytokine score (A, n=9 for each group). This differentiation is independent from the patients’ risk profile (B, definition for high risk situations see section patients and methods).